 recently, sarah friar, chief financial officer of openai, published an extensive article that systematically traces chatgpt’s evolution—from a curiosity‑arousing tool launched as a “research preview” three years ago to a foundational infrastructure supporting the daily work and lives of hundreds of millions of people worldwide. for the first time, she unveiled openai’s business logic—and its growth data—centered on computational power as the core constraint, with “intelligence value” serving as the pricing anchor.
the article points out that when chatgpt was first introduced, the goal was simply to observe what would happen when cutting‑edge intelligence was put directly into the hands of ordinary people. however, the subsequent wave of large‑scale adoption and deep, widespread use far exceeded everyone’s expectations. students used it to tackle late‑night homework challenges; parents relied on it to plan trips and manage family budgets; writers turned to it to break through creative blocks; and more and more individuals began using it to sort through health symptoms, prepare for doctor’s visits, and weigh complex decisions—turning to it in moments of fatigue, anxiety, or uncertainty to help them “think more clearly.”
this personal-level “boost” quickly found its way into professional settings: from polishing a document before meetings and double‑checking spreadsheets to rewriting a customer email to fine‑tune tone and phrasing; from engineers accelerating code inference to marketing teams refining their campaigns, from finance teams modeling scenarios with greater clarity to executives crafting more thoughtful, well‑prepared communications—chatgpt rapidly embedded itself into everyday workflows, becoming a tool that helps users “do more, decide faster.”
sarah friar emphasized that this transformation lies at the heart of openai’s self‑definition: as a company that balances “research with deployment,” its mission is to continuously narrow the gap between “advances at the forefront of intelligence” and “the actual adoption and utilization by individuals, businesses, and even nations.” as chatgpt became a “truly indispensable tool for getting things done,” openai adhered to a principle often described as “simple yet enduring”: a business model must expand in tandem with the value created by intelligence.
guided by this principle, openai steadily advanced its commercialization strategy. as user demand for capabilities and reliability grew, the company launched consumer‑oriented subscription services; as artificial intelligence delved deeper into teams and workflows, it developed workplace‑focused subscription products and introduced usage‑based pricing, aligning costs with “the actual volume of work completed.” building on this foundation, openai cultivated a platform‑driven business model, enabling developers and enterprises to embed intelligence into their own products and systems via apis—so that spending becomes directly proportional to real output.
more recently, this same logic has been extended to the realms of commerce and advertising: people turn to chatgpt not only to ask questions but also to decide their next “what to do”—what to buy, where to go, which option to choose. at the critical juncture where users shift from “exploration” to “action,” the platform both creates value for users and partners and provides space for advertising—after all, relevant options are truly valuable only when they are “clearly labeled and genuinely useful” as users approach decision‑making. across every path, openai maintains a unified standard: monetization methods must be “natively integrated into the user experience”; any commercialization that fails to add value “should have no place.”
in terms of usage data, openai revealed that chatgpt’s weekly active users (wau) and daily active users (dau) have consistently reached new all‑time highs. this growth stems from a flywheel that spans “computational power, cutting‑edge research, product development, and commercialization”: investments in computational power drive leaps in cutting‑edge research and model capabilities; stronger models deliver better product experiences and wider platform adoption; adoption generates revenue, which in turn fuels the next round of investment in computing power and innovation—creating a cycle of compounding growth.
looking back over the past three years, openai has used revenue as a key metric to gauge its “ability to serve customers,” noting that this metric has nearly mirrored the trajectory of available computational power. from 2023 to 2025, computational power saw an average annual increase of roughly three times, with overall growth reaching about 9.5 times: starting from 0.2 gigawatts in 2023, it rose to 0.6 gigawatts in 2024, then to approximately 1.9 gigawatts in 2025. the revenue curve closely followed this trajectory, with an average annual growth rate of around three times during the same period—rising from $2 billion in annual recurring revenue in 2023 to $6 billion in 2024, and then surpassing $20 billion in 2025, achieving roughly a tenfold expansion. sarah friar described this as “growth that is scalable at such a massive scale,” adding that if openai could maintain its current pace, the potential for further growth would be immense.
in her view, “computational power is the most scarce resource in today’s artificial intelligence landscape.” three years ago, openai relied almost entirely on a single provider of computational power; today, however, the company collaborates with “multiple suppliers within a diversified ecosystem.” this shift has brought resilience—and, more importantly, “certainty in computational power”: in a market environment where “the ability to secure computational power directly determines who can scale,” openai can now plan, finance, and deploy capacity with greater confidence.
under this strategy, computing power shifts from being a “fixed constraint” to an “actively managed portfolio.” when training cutting-edge models, openai leverages top-tier hardware during the most critical stages of development; when serving large-scale inference workloads, it places greater emphasis on cost-effectiveness, deploying lower-cost infrastructure that prioritizes efficiency over sheer scale. the result is reduced latency, increased throughput, and the ability to deliver “intelligently useful” models at a price point of just a few cents per million tokens—making artificial intelligence accessible for everyday workflows rather than confined to a handful of “elite” use cases.
building on top of its computing power, openai has constructed an integrated product platform spanning text, images, speech, code, and apis—empowering individuals and organizations to think, create, and operate more efficiently. the next phase will focus on “intelligent agents and workflow automation”: these systems will run continuously, maintain context over time, and take action across multiple tools—meaning that for individuals, ai will manage projects, coordinate schedules, and execute tasks; for organizations, it will evolve into the “operating layer of knowledge work.” as these systems transition from “novelty” to “habit,” their usage will become deeper and more enduring—and this newfound predictability will strengthen platform economics, underpinning long-term investment.
in terms of its business model, openai initially launched with a subscription-based approach, but today it operates a multi-tiered system: at one end are paid subscriptions for consumers and teams, while at the other lies a free tier supported by advertising and commercial partnerships to drive mass adoption; in between sits a pay‑per‑use api service tightly integrated with production environments. looking ahead, as intelligence permeates fields such as scientific research, drug discovery, energy systems, and financial modeling, new economic models will continue to emerge—including licensing partnerships, intellectual property–based agreements, and outcome‑based pricing models that enable value sharing. sarah friar notes that this trajectory closely mirrors the evolution of the internet industry—and that “intelligence will follow the same path forward.”
this framework places higher demands on “discipline.” in her words, securing world‑class computing power often requires making commitments years in advance, yet business growth is rarely a perfectly smooth straight line: sometimes capacity outstrips demand, while at other times demand surges ahead of supply. openai balances this tension by maintaining a “light‑asset” balance sheet, replacing full-stack ownership with strategic partnerships, and preserving contractual flexibility across diverse vendors and hardware types. capital is allocated in batches according to “real demand signals”—a structure that allows the company to proactively “lean forward” when growth arrives, while avoiding the premature locking in of excessive resources for future needs that have yet to be validated by the market.
guided by this “discipline,” openai has set its core focus for 2026 on “pragmatic adoption.” its top priority is to narrow the gap between what “artificial intelligence can already do today” and what “individuals, businesses, and nations actually use in their daily lives.” she emphasizes that this opportunity is “massive and urgent,” especially in healthcare, science, and enterprise settings, where better intelligence can almost directly translate into improved outcomes.
at the conclusion of her article, sarah friar outlines openai’s understanding of “scaling intelligence” through a set of interrelated concepts: infrastructure determines “what we can deliver,” innovation defines “what intelligence can do,” adoption decides “who gets to use it,” and revenue funds “the next leap forward.” in her view, this is precisely the pathway by which intelligence continues to expand—and ultimately becomes the “foundation of the global economy.” |
